Trango ComputeContextIQ
Personal Agent ProtocolPAPPoppypoppy.jsonSierraMetaOAuth 2.0MCPAI agentsexplainer

What Is the Personal Agent Protocol (PAP)? A Plain Explanation of Sierra and Meta's Standard for AI Agents and Businesses

Personal Agent Protocol (Poppy) explained simply: what PAP is, the poppy.json file, how guest and signed-in agent sessions work with OAuth, and who announced it: Sierra, Meta, Shopify, Stripe, Walmart.

October 6, 2026Trango Compute Inc.

The Personal Agent Protocol (PAP, also called Poppy) is an open standard that defines how a person's AI agent connects to a business and what the business lets it do. Sierra announced it on October 6, 2026, with Meta co-developing and Genesys, Instinct, Rocket, Shopify, Stripe and Walmart named as partners.

Updated October 9, 2026. The Draft 0.1 specification is now published at personalagentprotocol.org. This post has been rewritten to match it. It is a draft, and the site says any part of it can still change.

The short version

Think of a personal AI agent as an assistant you send to a store on your behalf. Today it has to walk in the front door like any customer: load web pages, find buttons, fill in forms. Often a company either blocks it or the agent signs in with your password. PAP proposes a standard side entrance, so the agent and the business can exchange what they need directly, and so you and the company can both choose how much the agent may do.

How a PAP session works

The draft specification describes it in four steps:

  1. Discover. A company publishes a small file at /.well-known/poppy.json that says who it is and how agents can work with it: sign-in, APIs, a website, or its own company agent.
  2. Start a session. The agent proves who it is by signing a request with its own key. As announced, it can start as a guest for simple questions like stock or returns policy, and the session records whether a user has signed in.
  3. The customer signs in. For anything account-specific, the customer signs in. The company grants scopes: poppy:read to view, poppy:write to change, or custom scopes it defines. There are three ways to sign in: in the customer's browser, on another device with a code, or by handing the agent credentials directly (the least preferred).
  4. The session continues. One session spans API calls, conversations and web browsing, so what happened before and after sign-in stays connected.

Session tokens are tied to a key the agent holds, using a standard called DPoP. A stolen token alone does not work.

How a business can connect

A company picks whichever routes suit it, and can add more later:

  • Its website. The agent joins its browser to the session.
  • An API. Described with MCP or OpenAPI.
  • Its own agent. For conversational tasks like a warranty claim, with a way to hand off to a person.

Companies also decide what agents may do. The project's homepage describes five permission levels, from blocking agents entirely to full autonomy.

Is PAP a payment protocol?

No. Payments are listed as an open topic: "how a personal agent makes payments in a way that fits the protocol, including how that works with existing payment protocols." Nothing is specified yet.

If you want the payment side, read about x402, which lets an agent pay an HTTP endpoint. AP2 (Google's Agent Payments Protocol) is a separate effort for proving a user authorized a purchase.

What is still unknown

  • Stability. It is Draft 0.1 and can still change.
  • Governance. The specification and documentation use the Apache License 2.0, which excludes the names "Personal Agent Protocol" and "Poppy". The site names no foundation or steward.
  • Open topics. Payments, push notifications and attachments are not specified yet.
  • Real-world use. We found no reported production deployments.

For a deeper comparison with MCP, AP2 and x402, and a table of what the draft defines, see Personal Agent Protocol Explained.

Why it matters if you run an API or MCP server

PAP sits above your existing integrations rather than replacing them. If you already expose an MCP server or an OpenAPI description, you are on one of the routes. The MCP Inspector shows what an agent can discover about your server today, and the Poppy Inspector reads and lints a poppy.json and cross-checks its OAuth issuer (it does not call the endpoints the file lists). A clean OAuth setup is the most reusable preparation, and the OIDC Inspector shows what your authorization server publishes.

Sources

Try ContextIQ free

Free tools for AI engineers.

Follow Trango Compute on LinkedIn

We post updates on new tools, context engineering patterns, and LLM cost research.

Follow on LinkedIn